• Summary
• Related Documents
  • Active Bibliography
  • Co-citation
• Version History

Abstract:

We propose to develop a portable middleware service to manage dynamic allocation and scheduling of computing resources in embedded distributed real-time systems transparently on behalf of the application programmer. The service ensures satisfaction of timing constraints subject to given load and failure assumptions in accordance with application Quality of Service (QoS) requirements while achieving graceful QoS adaptation when the load or failure hypothesis is violated. Traditionally pre-run-time feasibility analysis is employed to ensure that timing constraints of embedded real-time applications are satisfied at run-time. This analysis requires a priori characterization of run-time load and failure conditions. As embedded systems grow more complex, characterization of the worst-case conditions becomes more difficult. Thus, such systems must be designed to deal with potential violations of the load and failure hypotheses. We argue that current priority-based systems and dynamic admission control based systems do not allow for graceful degradation because they do not give the designer fine-grained means to express how exactly application QoS should be degraded. Instead, we propose a QoS adaptation model based on QoS negotiation with the application as a means to meet both predictability and graceful degradation requirements. Our model provides generic means for the

Citations